What does citation presence measure?
Whether a site appears as a cited source when an engine answers the questions its buyers ask, measured on a fixed prompt set the manager approved, sampled by a Search Visibility Digital FTE on ftes.ai every week across the search and AI engines the manager selected, with the sample date recorded. It is the primary outcome measure for the search visibility role and the honest replacement for a ranking promise, which no one can make. It is reported per engine, because engines cite differently, and as a range per cycle, because engine answers vary between samples.
The myth piece on guaranteed AI visibility explains why a range with dates is a measurement and a promise is not.
How is it computed, with the inputs left blank?
Prompt set: [N] questions, approved on [date], version [version]. Engines: [engine], [engine], [engine]. For each weekly sample on [date]: for each engine, the count of prompts whose answer cited [site], [C], divided by [N], giving that week's presence for that engine, [rate]. Over the cycle from [date] to [date] with [W] weekly samples: per engine, the lowest and highest weekly presence, [rate] to [rate], reported as the range, with the median, [rate], if useful. Baseline: the first sample, on [date], before any page was published, [rate] per engine. Companion: console impressions and clicks for the target queries, [figures], where the console is connected.
A new prompt set or a new engine starts a new baseline and is labelled; the old series is kept for reference and not compared.
- Per week, per engine: cited prompts over the set
- Per cycle, per engine: the range across weekly samples
- Baseline: the first sample before publishing
Why a range and not a number?
Because the same question asked of the same engine on two days can produce different answers with different citations, and a single sample would report noise as movement. The weekly cadence with the date recorded produces a series; the range over the cycle shows both where the site sits and how much the engine varies. Movement is read as the range shifting across cycles, not as one week beating another. A vendor reporting a single citation figure for a month is reporting one sample, and the manager should ask which one.
The whitepaper on the AI search visibility playbook describes the sampling design in full.
What is unmeasured, and why is it labelled?
Reach on platforms the FTE cannot sample or connect to; attribution of an enquiry or sale to a page, which no platform can see honestly; satisfaction of the reader; and any outcome on an engine outside the selected set. The performance view marks each unmeasured rather than estimating it, and the monthly report repeats the labels. This is what makes the measured numbers credible: a manager who sees the platform say unmeasured about three things believes the twelve it measured. The agency reporting piece shows the same principle from the client's side.
Anything a vendor claims to measure in this list is being estimated, and the manager should ask how.
How does it mislead?
When the prompt set is changed to include questions the site already wins, which raises the number and measures nothing; the set is versioned so this is visible. When a single week is compared with another single week; compare ranges across cycles. When a rise on one engine is reported as a rise overall; report per engine. When movement in the first month is read as a result; visibility compounds over a quarter, and the first cycle is the baseline being established. And when console movement is attributed entirely to the FTE's pages; the console shows the queries, and the pages published in the period are listed beside them, but the attribution is stated as an association.
The worth-it verdict uses this same sampling as its first question, so the measure and the topic selection share a source.
How is it applied?
Approve the prompt set and the engines before any page is written, and let the FTE take the baseline sample. Read the weekly samples in the Performance tab with the pages published that week. At the monthly review, write the per-engine range against the previous cycle and the baseline, list what is unmeasured, and decide next cycle's cluster from the verdicts. Do not draw a conclusion a CMO would repeat until the quarterly review, and then draw it as a range with dates.
The ninety-day search visibility playbook schedules the samples and the reviews.